acceptodds
Withdrawn from ICLR 2027

Directly Optimizing Mean Demographic Parity for Nonlinear Regression

Withdrawn from ICLR 2027, so left out of the paper lists and the map.

Abstract

We focus on regression settings where the fairness goal is to equalize average predictions across values of a sensitive attribute, a criterion known as mean demographic parity. Directly optimizing this criterion is difficult because it depends on a conditional mean that is unknown and changes during training. Common dependence penalties and adversarial methods do not estimate this conditional mean; instead, they push predictions toward full independence. This stronger constraint can reduce accuracy even when average predictions are already equal. Existing conditional-mean methods are limited to linear predictors or low-dimensional sensitive attributes. We enable direct optimization of mean demographic parity using DPVar, a fairness measure defined as the variance of the conditional mean prediction. Because the conditional mean must be estimated as the predictor changes, optimizing DPVar leads to a functional bilevel problem. We develop two solvers: FBO, which uses a closed-form hypergradient, and an iterative-differentiation (ITD) solver that differentiates through updates of the conditional-mean estimator. Unlike previous conditional-mean methods, our approach applies to nonlinear predictors and high-dimensional continuous sensitive attributes. Across a semi-synthetic benchmark built from 21 tabular regression datasets and Communities & Crime data, FBO and ITD recover competitive or better accuracy–DPVar trade-offs than existing methods.

Primary area:
alignment, fairness, safety, privacy, and societal considerations
Keywords:
Algorithmic fairness, mean demographic parity, fair regression, bilevel optimization, conditional-mean estimation.

Reviews

Among:
Shape:
1234this paper, Bottom 10%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.67 over 3 reviews, above 10% of the others.
Reviewer_ssT1OpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
2
Soundness
2
Read review
Paper summary

The paper studies mean demographic parity in regression when the sensitive attribute is continuous and potentially multidimensional. It uses DPVar, defined as the variance of the conditional mean E[f(X)∣A]\mathbb{E}[f(X)\mid A] across the sensitive attribute AA, to measure fairness. The authors formulate the problem as functional bilevel optimization, where an inner neural network estimates the conditional mean and an outer network minimizes the prediction loss together with a DPVar penalty. Two optimization approaches are proposed: FBO, which uses an explicit hypergradient, and ITD, which differentiates through the unrolled inner updates. The experiments include a controlled simulation, a 21-task semi-synthetic benchmark based on OpenML datasets, and the Communities & Crime (C&C) dataset.

Critical strengths
  1. Mean demographic parity for continuous and multidimensional sensitive attributes is a relevant problem. The paper also makes a clear distinction between mean parity and full statistical independence, which helps motivate the proposed fairness criterion.

  2. The hypergradient derivation appears correct, and the resulting FBO update is relatively simple and computationally efficient.

  3. The controlled simulation includes an analytic oracle, which is useful for evaluating the method. The authors have also released their code, making the experiments easier to reproduce.

Critical weaknesses
  1. The novelty is limited, and the paper does not sufficiently discuss closely related work. Mean-parity regression, fairness with continuous sensitive attributes, and bilevel optimization have all been studied in previous work. Several papers that seem particularly relevant are missing from the discussion, making it difficult to identify the main methodological contribution.

    For mean parity and conditional-mean fairness in regression, relevant papers include:

    • Komiyama et al., "Nonconvex Optimization for Regression with Fairness Constraints" (ICML 2018).
    • Fitzsimons et al., "A General Framework for Fair Regression" (Entropy 2019).
    • Scutari et al., "Achieving Fairness with a Simple Ridge Penalty" (Statistics and Computing 2022).
    • Wei et al., "Mean Parity Fair Regression in RKHS" (AISTATS 2023).
    • Giuliani et al., "Generalized Disparate Impact for Configurable Fairness Solutions in ML" (ICML 2023). In particular, this work already considers conditional-mean fairness for continuous protected attributes.

    More importantly, the DPVar objective can be written as

�0�

.$$

This gives a min-max formulation in which an adversary tries to predict the model output from the sensitive attribute. From this perspective, FBO appears closely related to alternating gradient descent-ascent. Similar ideas have already been explored in Zhang et al., "Mitigating Unwanted Biases with Adversarial Learning" (AIES 2018), and in neural HGR-based fairness methods such as Grari et al., "Fairness-Aware Neural Rényi Minimization for Continuous Features" (IJCAI 2020).. I think the authors need to explain more carefully what is new beyond applying these existing techniques to the DPVar objective.

  1. The baselines are weak, and some comparisons are not fair. I have several concerns about the experimental comparisons.

    • GDP uses a restricted representation of the sensitive attribute. The baseline is applied to a two-dimensional PCA projection of AA. In the controlled experiment, this projection removes the only direction responsible for the disparity, putting GDP at a clear disadvantage.

    • The HSIC baseline does not use the same model capacity. It is implemented with a linear predictor and a linear kernel, whereas FBO and ITD use neural networks. As a result, the comparison mixes differences in model expressiveness with differences in fairness regularization. It is difficult to determine whether the reported improvements actually come from the proposed fairness method.

    • Several important competitors are missing, which include HGR-NN (cited but not evaluated), nonlinear-kernel HSIC and MMD penalties, GeDI, and reduction-based approaches for fair regression. These methods would provide much stronger comparisons, particularly if all methods used the same downstream predictor architecture.

    • The proposed methods do not consistently outperform even the weaker baselines. On the C&C dataset, the linear HSIC baseline matches or outperforms FBO and ITD at most DPVar budgets, according to the authors' released aggregate_frontiers.csv. This makes the claimed empirical advantage less convincing.

  2. The DPVar estimator itself is not adequately validated. This is one of my main concerns. In both the tabular benchmark and the C&C experiments, fairness is evaluated using a neural DPVar estimator from the same model family as the proposed method. However, it is unclear whether this estimator can reliably detect nonlinear dependence between the predictions and sensitive attributes.

    The authors' released controlled simulation results (summary.csv) illustrate this problem. In one setting, the estimated DPVar is only 0.0700.070, while the true value is 0.9340.934. Along the FBO optimization path, the true DPVar decreases by approximately 80-fold, whereas the estimated DPVar decreases by only about threefold.

    This is a substantial discrepancy. It suggests that the estimator may miss much of the nonlinear disparity that the proposed method is intended to address. Since the same type of estimator is used to evaluate fairness in the real-data experiments, I am concerned that the reported fairness improvements may not accurately reflect the true level of disparity.

    I think an independent fairness evaluation, preferably using a stronger and more flexible estimator, is needed before drawing conclusions about the effectiveness of the proposed methods.

Questions for authors
  1. Beyond applying functional bilevel optimization to DPVar, what are the main methodological contributions relative to HGR-NN, GeDI, and Wei et al., "Mean Parity Fair Regression in RKHS" (AISTATS 2023)? Also, do the authors agree that the proposed FBO method can be interpreted as alternating gradient descent-ascent on the equivalent min-max problem? If so, what is the advantage of the bilevel formulation?

  2. How do FBO and ITD compare with HGR-NN, nonlinear-kernel HSIC/MMD penalties, and full-dimensional GDP when all methods use the same neural predictor architecture and comparable training budgets?

  3. Could the authors validate the DPVar estimates using an independent, high-capacity estimator, particularly in the controlled setting where the true DPVar is available? I would also like to see ablation studies on the number of inner steps, unrolling depth, and inner-network capacity to understand how these choices affect the results.

Reviewer_q9QZOpenReview ↗
Rating
3
Confidence
3
Informativeness
2
Presentation
3
Soundness
3
Read review
Paper summary

This paper proposes to enforce mean demographic parity for continuous sensitive features where the conditional mean predictions should be as constant as possible with respect to the sensitive attributes. For this it penalizes during training the DPVar criterion which corresponds to the variance of the conditional mean prediction with an hyperparameter trade-off. For penalizing DPVar it proposes to regress the prediction itself on the sensitive attribute and considers two methods to compute the hypergradient: one with a closed-form expression FBO, and another by differentiating through several steps of gradient descent, ITD. Experiments are done on semi synthetic datasets and on Communities & Crime with a ten-dimensional sensitive attribute and show the effectiveness of the methods in comparison to baselines.

Critical strengths

The paper is clear, easy to follow and I found the FBO approach well grounded: hypergradient has a simple closed form with no Hessian inversion. Empirically the two methods show the strongest win rates on 21 datasets.

Critical weaknesses
  1. Baseline choices limit the comparison, the HSIC baseline uses linear kernels, which cannot detect nonlinear dependence. I was expecting an RBF kernel baseline, since approximations (e.g. random features, Nyström) could reduce its cost. Also the adversarial method does not specify what model structure is used, and no HGR or Mutual Information penalization/adversarial is compared (HGR baselines in the literature already show reductions of quantile-based mean parity metrics and can take multidimensional features).
  2. Training algorithms use separated IN and OUT splits, each term sees only one part of the training data, if too small the terms can be poorly estimated.
Questions for authors
  1. Could you add other baselines such as RBF kernel HSIC and HGR or mutual-information penalization? Could you also specify the architecture of the adversary?
  2. Why not use separate minibatches instead of fixed IN/OUT splits?
  3. Could you share the fit quality of hϕh_ϕ throughout training?
  4. In the FBO penalty, hϕh_ϕ is fitted on the IN samples and the fairness term is also averaged over the IN samples while ITD evaluates the variance over the OUT samples. Is this intentional, and does it affect the FBO gradient?
Reviewer_w8Z8OpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

This paper studies mean demographic parity in nonlinear regression with continuous, multivariate sensitive attributes. The objective combine least squared error with DPVar, which is the variance of conditional mean prediction given the sensitive attributes. The authors use the conditional mean as the solution of an inner least-squares regression. The FBO solver uses a closed-form population hypergradient and ITD solver differentiates through a finite sequence of inner updates.

Critical strengths
  1. This paper proposes an interesting fairness criterion that targets the average prediction disparities, and extend it to nonlinear predictors and multidimensional continuous sensitive attributes without discretization or kernal smoothing.

  2. The functional bilevel approach links the conditional-mean estimation with predictive optimization. For squared-loss inner regression. explicit population hypergradient allows FBO updates to skip Hessian inversion.

Critical weaknesses
  1. DPVar only control mean prediction disparities, and leave variance, tail, and the distributional features unconstrained, and thus do not guarantee equal opportunity for individual fairness.

  2. While training and evaluation require estimating the unknown conditional-mean function, this can be extremely difficult as dimensionality grows substantially. Moreover, results from Communities & Crime do not support individual-level or causal fairness claims.

Questions for authors
  1. The paper integrates DPVar into a least-squares objective as a penalty to encourage approximate mean demographic parity in regression. However, the methodological contribution appears incremental, and the manuscript does not sufficiently establish a substantive advance over existing mean-parity methods and functional bilevel optimization approaches.

  2. Given that the empirical evaluation relies primarily on the regularization of DPVar, would the reported advantages of FBO and ITD persist under other fairness measures, such as absolute conditional-mean disparities or distributional dependence? These comparisons would help clarify the robustness and scope of the reported fairness improvements.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.